Construction of web - based prediction nomogram models for cancer - specific survival in patients at stage IV of hepatocellular carcinoma depending on SEER database

Abstract

Objectives: Hepatocellular carcinoma (HCC) prognosis involves multiple clinical factors. Although nomogram models targeting various clinical factors have been reported in early and locally advanced HCC, there are currently few studies on complete and effective prognostic nomogram models for stage IV HCC patients. This study aims to creat nomograms for cancer-specific survival (CSS) in patients at stage IV of HCC and developing a web predictive nomogram model to predict patient prognosis and guide individualized treatment.

Methods: Clinicopathological information on stage IV of HCC between January, 2010 and December, 2015 was collected from the Surveillance, Epidemiology, and End Results (SEER) database. The patients at stage IV of HCC were categorized into IVA (without distant metastases) and IVB (with distant metastases) subgroups based on the presence of distant metastasis, and then the patients from both IVA and IVB subgroups were randomly divided into the training and validation cohorts in a 7꞉3 ratio. Univariate and multivariate Cox regression analyses were used to analyze the independent risk factors that significantly affected CSS in the training cohort, and constructed nomogram models separately for stage IVA and stage IVB patients based on relevant independent risk factors. Two nomogram's accuracy and discrimination were evaluated by receiver operator characteristic (ROC) curves and calibration curves. Furthermore, web-based nomogram models were developed specifically for stage IVA and stage IVB HCC patients by R software. A decision analysis curve (DCA) was used to evaluate the clinical utility of the web-based nomogram models.

Results: A total of 3 060 patients were included in this study, of which 883 were in stage IVA, and 2 177 were in stage IVB. Based on multivariate analysis results, tumor size, alpha-fetoprotein (AFP), T stage, histological grade, surgery, radiotherapy, and chemotherapy were independent prognostic factors for patients with stage IVA of HCC; and tumor size, AFP, T stage, N stage, histological grade, lung metastasis, surgery, radiotherapy, and chemotherapy were independent prognostic factors for patients with stage IVB HCC. In stage IVA patients, the 3-, 6-, 9-, 12-, 15-, and 18-month areas under the ROC curves for the training cohort were 0.823, 0.800, 0.772, 0.784, 0.784, and 0.786, respectively; and the 3-, 6-, 9-, 12-, 15-, and 18-month areas under the ROC curves for the validation cohort were 0.793, 0.764, 0.739, 0.773, 0.798, and 0.799, respectively. In stage IVB patients, the 3-, 6-, 9-, and 12-month areas under the ROC curves for the training cohort were 0.756, 0.750, 0.755, and 0.743, respectively; and the 3-, 6-, 9-, and 12-month areas under the ROC curves for the validation cohort were 0.744, 0.747, 0.775, and 0.779, respectively; showing that the nomograms had an excellent predictive ability. The calibration curves showed a good consistency between the predictions and actual observations.

Conclusions: Predictive nomogram models for CSS in stage IVA and IVB HCC patients are developed and validated based on the SEER database, which might be used for clinicians to predict the prognosis, implement individualized treatment, and follow up those patients.

Keywords: SEER database; cancer-specific survival; hepatocellular carcinoma; nomogram.

Figure 1. Flow chart of patient’s screening

Figure 2. Nomogram of CSS in stage IVA HCC patients at 3-, 6-, 9-, 12-, 15-, and 18-month

*P<0.05, ***P<0.001. CSS: Cancer-specific survival; IVA: Stage IV without distant metastases; HCC: Hepatocellular carcinoma.

Figure 3. Nomogram of CSS in stage IVB HCC patients at 3-, 6-, 9-, and 12-month

***P<0.001. CSS: Cancer-specific survival; IVB: Stage IV with distant metastases; HCC: Hepatocellular carcinoma.

Figure 4. ROC curves of the nomogram

A and B: AUCs for predicting CSS in training cohort (A) and validation cohort (B) of stage IVA HCC patients; C and D: AUCs for predicting CSS in training cohort (C) and validation cohort (D) of stage IVB HCC patients. ROC: Receiver operator characteristic; AUC: Area under the curve; CSS: Cancer-specific survival; IVA: Stage IV without distant metastases; IVB: Stage IV with distant metastases; HCC: Hepatocellular carcinoma.

Figure 5. Calibration curves of the nomogram

A and B: Calibration curves of the 3-, 6-, 9-, 12-, 15-, and 18- month predicting CSS in training cohort (A) and validation cohort (B) of stage IVA HCC patients; C and D: Calibration curves of the 3-, 6-, 9-, and 12- month predicting CSS in training cohort (C) and validation cohort (D) of stage IVB HCC patients. CSS: Cancer-specific survival; IVA: Stage IV without distant metastases; IVB: Stage IV with distant metastases; HCC: Hepatocellular carcinoma.

Figure 6. DCA curves of the nomogram

A and B: DCA curves of the 3-, 6-, 9-, 12-, 15-, and 18- month predicting CSS in training cohort (A) and validation cohort (B) of stage IVA HCC patients; C and D: the calibration curves of the 3-, 6-, 9-, and 12- month predicting CSS in training cohort (C) and validation cohort (D) of stage IVB HCC patients. DCA: Decision curve analysis; CSS: Cancer-specific survival; IVA: Stage IV without distant metastases; IVB: Stage IV with distant metastases; HCC: Hepatocellular carcinoma.

Figure 7. A web nomogram for predicting CSS for patients with stage IVA HCC

CSS: Cancer-specific survival; IVA: Stage IV without distant metastases; HCC: Hepatocellular carcinoma.